A Conserved Sequence Immediately N-terminal to the Bateman Domains in AMP-activated Protein Kinase γ Subunits Is Required for the Interaction with the β Subunits*

Mammalian AMP-activated protein kinase is a serine/threonine protein kinase that acts as a sensor of cellular energy status. AMP-activated protein kinase is a heterotrimer of three different subunits, i.e. α, β, and γ, with α being the catalytic subunit and β and γ having regulatory roles. Although several studies have defined different domains in α and β involved in the interaction with the other subunits of the complex, little is known about the regions of the γ subunits involved in these interactions. To study this, we have made sequential deletions from the N termini of the γ subunit isoforms and studied the interactions with α and β subunits, both by two-hybrid analysis and by co-immunoprecipitation. Our results suggest that a conserved region of 20–25 amino acids in γ1, γ2, and γ3, immediately N-terminal to the Bateman domains, is required for the formation of a functional, active αβγ complex. This region is required for the interaction with the β subunits. The interaction between the α and γ subunits does not require this region and occurs instead within the Bateman domains of the γ subunit, although the α-γ interaction does appear to stabilize the β-γ interaction. In addition, sequential deletions from the C termini of the γ subunits indicate that deletion of any of the CBS (cystathionine β-synthase) motifs prevents the formation of a functional complex with the α and β subunits.

Mammalian AMP-activated protein kinase is a serine/threonine protein kinase that acts as a sensor of cellular energy status. AMP-activated protein kinase is a heterotrimer of three different subunits, i.e. ␣, ␤, and ␥, with ␣ being the catalytic subunit and ␤ and ␥ having regulatory roles. Although several studies have defined different domains in ␣ and ␤ involved in the interaction with the other subunits of the complex, little is known about the regions of the ␥ subunits involved in these interactions. To study this, we have made sequential deletions from the N termini of the ␥ subunit isoforms and studied the interactions with ␣ and ␤ subunits, both by two-hybrid analysis and by coimmunoprecipitation. Our results suggest that a conserved region of 20 -25 amino acids in ␥1, ␥2, and ␥3, immediately N-terminal to the Bateman domains, is required for the formation of a functional, active ␣␤␥ complex. This region is required for the interaction with the ␤ subunits. The interaction between the ␣ and ␥ subunits does not require this region and occurs instead within the Bateman domains of the ␥ subunit, although the ␣-␥ interaction does appear to stabilize the ␤-␥ interaction. In addition, sequential deletions from the C termini of the ␥ subunits indicate that deletion of any of the CBS (cystathionine ␤-synthase) motifs prevents the formation of a functional complex with the ␣ and ␤ subunits.
Mammalian AMP-activated protein kinase (AMPK) 2 is a serine/threonine protein kinase that acts as a sensor of cellular energy status. It is activated by cellular stresses that deplete ATP, either by inhibiting ATP production (e.g. hypoxia, glucose deprivation, heat shock, and mitochondrial inhibitors) or by accelerating ATP consumption (e.g. muscle contraction). Depletion of ATP is always accompanied by increases in AMP due to the reaction catalyzed by adenylate kinase, and the increase in AMP:ATP ratio activates AMPK in an ultrasensitive manner (1). Once activated it switches on catabolic pathways and switches off many ATP-consuming processes, including anabolic pathways (see Refs. 2-5, for reviews). AMPK is a heterotrimer composed of three different subunits, i.e. ␣, ␤, and ␥. The ␣ subunit is the catalytic subunit; it contains a highly conserved kinase domain at the N terminus and a less well conserved C-terminal regulatory domain. Two isoforms have been described, i.e. ␣1 and ␣2; both are localized in the cytoplasm, although ␣2 is also present in the nucleus (6). The ␥ subunits contain four tandem repeats of a structural module called a CBS motif (7), named after the enzyme cystathionine ␤-synthase, in which a pair of CBS motifs form a domain that binds the allosteric activator S-adenosyl methionine (8). In the AMPK ␥ subunits the four CBS motifs are now known to act in two pairs, forming two domains (referred to as Bateman domains) that bind the regulatory nucleotides, AMP and ATP, in a mutually exclusive manner (8). Three isoforms of the ␥ subunit, i.e. ␥1, ␥2, and ␥3, are encoded by distinct genes; they have poorly conserved N-terminal regions that in ␥2 and ␥3 are subject to alternate splicing, whereas the C-terminal regions, containing the two tandem Bateman domains, are conserved in all three isoforms. Two isoforms of the ␤ subunit (␤1 and ␤2) have been described; they differ at their N termini, but both appear to interact with the ␣ and ␥ subunits with similar efficiency (9,10). The ␤ subunits have two conserved regions, a central glycogenbinding domain (11)(12)(13) and a C-terminal domain that is the only region required for the formation of the complex with ␣ and ␥ (11).
All three subunits are required to form a functional AMPK complex (14,15). Recent studies have begun to delineate the regions of the ␣ and ␤ subunits required for the formation of heterotrimeric complexes, although there have been some con-flicting findings. Detailed mapping of the C-terminal domain of ␤1 suggested that the last 25 residues are sufficient to form a complex with ␥1, ␥2, and ␥3, with the C-terminal residue (Ile 270 ) being essential for the formation of a ␤-␥ complex in the absence of ␣ (16). By contrast, it was recently reported that ␤2 and ␥1 do not form a complex in the absence of ␣2 (17). The C-terminal domain of the ␣ subunit has been reported to be involved in the interaction with ␤ (16) and ␥ subunits (10), although another study has suggested that ␥1 interacts with ␣2 via both the N-terminal catalytic domain and the C-terminal domain of the latter (17).
As yet, little is known about which portions of the ␥ subunits are involved in the interaction with the ␣ and ␤ subunits. To study this, we have made deletions of the regions of the ␥ subunits N-terminal to the Bateman domains. Our results show that a conserved sequence of 20 -25 residues, immediately N-terminal to the first Bateman domain (the "pre-CBS1 sequence"), is required for the binding of ␥1, ␥2, and ␥3 to the ␤ subunits and for the formation of an active ␣␤␥ complex. In addition, we have made C-terminal deletions of the ␥ subunits and observed that deletion of any of the CBS motifs prevents the formation of functional complexes with the ␣ and ␤ subunits.

MATERIALS AND METHODS
Microorganisms, Culture Conditions, and Genetic Methods-Escherichia coli DH5␣ was used as the host strain for plasmid constructions. It was grown in LB (1% peptone, 0.5% yeast extract, 1% NaCl, pH 7.5) medium supplemented with 50 mg/liter ampicillin.
Oligonucleotides-Oligonucleotides used in this work are described in Table 1.
␤-Galactosidase Assays-␤-Galactosidase activity was measured in yeast-permeabilized cells and expressed in Miller units as described by Ludin and collaborators (24).
Preparation of Yeast Cell Extracts and Immunoblot Analysis-Yeast cells corresponding to 1 unit of A 600 were collected by rapid centrifugation (14,000 rpm, 1 min), resuspended in 100 l of Laemmli sample buffer, and boiled for 3 min. Glass beads (0.3 g, 450-m diameter) were added to the suspension, and then the cells were vortexed at full speed for 30 s. The suspension was boiled again for 3 min and centrifuged at 14,000 rpm for 1 min. 20 l of the supernatants was subjected to SDS-PAGE and immunoblotting using anti-LexA polyclonal antibodies (Invitrogen) or anti-HA polyclonal antibodies (Clontech).
Cell Culture and Transfection-HeLa and HEK293 cells were grown at 37°C in a 5% CO 2 incubator. Cells were grown in Dulbecco's modified Eagle's medium supplemented with 10% fetal bovine serum. All transfections were carried out using SuperFect transfection reagent (Qiagen) following the manufacturer's instructions.
Kinase Assays-Activity of AMPK in immunoprecipitates was measured as described previously (25).

RESULTS
The ␥ Subunits Interact with ␣ and ␤ via Their N-terminal Regions by Two-hybrid Analysis-We commenced our study by analyzing two-hybrid interactions of mammalian AMPK subunits expressed in yeast. Because the ␣2␤2␥3 complex appears to play an important role in skeletal muscle (27), we initially studied the interaction between ␤2 and ␥3, which has an N-terminal extension of 201 amino acids prior to the first Bateman domain. An LexA-␥3 fusion protein interacted strongly with a GAD-␤2 fusion (Fig. 1), so we made sequential deletions from the N terminus of ␥3 and checked their interaction with GAD-␤2. As shown in Fig. 1B, deletion of the first 33 amino acids of ␥3 (␥3A) decreased the interaction slightly, whereas further successive deletions of 45 (␥3B) and 40 (␥3C) amino acids did not further modify the interaction. However, the deletion of the following 35 amino acids (␥3D) reduced the interaction, and the deletion of the next 47 amino acids (␥3E, truncated at the start of the first Bateman domain) completely abolished it.

FIGURE 1. AMPK␥3 interacts with AMPK␤2 through its N-terminal region.
A, diagram of the deletions constructed. Deletions were made as described under "Materials and Methods" and fused in-frame to LexA. The box drawn with a dashed line indicates the region in AMPK␥3 involved in the interaction. B, two-hybrid analysis of the AMPK␥3-AMPK␤2 interaction. Yeast CTY10.5d strain was transformed with plasmids expressing GAD-AMPK␤2 and different truncated forms of LexA-AMPK␥3. Transformants growing exponentially in SC-4% glucose medium were harvested, and the ␤-galactosidase activity measured. Values correspond to means from four to six different transformants (Bars indicated Ϯ S.D.). C, Western blot analysis. Crude extracts from the transformants described in section B were analyzed by Western blot using anti-LexA polyclonal antibodies. One representative transformant from each interaction is shown.
Additional deletion of the first CBS motif (␥3F) gave the same negative results as ␥3E. Control experiments using the various LexA-␥3 constructs and the empty vector pACT2 gave negligible (Ͻ1 unit) ␤-galactosidase activity in all the cases (not shown). Western blot analysis indicated that all deleted forms were expressed at similar levels (Fig. 1C). These results suggested that the interaction between ␥3 and ␤2 requires the 47 amino acids immediately prior to the first Bateman domain, although residues N-terminal to that may improve the interaction. We also attempted to study the interaction between ␥3 and ␣2 using the same approach, but there was a very low level of interaction evident even with the full-length construct (Ͻ1 unit of ␤-galactosidase).
We next studied ␥1, the most abundant ␥ subunit in skeletal muscle (27). As shown in Fig. 2A, a fusion between LexA and full-length ␥1 was able to interact strongly with both ␣2 and ␤2. However, the deletion of the first 41 amino acids from the N terminus (a truncation at the start of the first Bateman domain) completely abolished the interaction with both ␣2 and ␤2. Control experiments using both LexA-␥1 constructs and the empty vector pACT2 gave negligible (Ͻ1 unit) ␤-galactosidase activity (not shown). Western blot analysis indicated that the truncated LexA-␥1 protein was produced at similar levels to the fulllength protein ( Fig. 2A, right panel).
Similar results were obtained when we studied ␥2. A fusion between LexA and full-length ␥2, containing an N-terminal extension of 278 amino acids prior to the first Bateman domain, interacted with ␣2 and ␤2 (Fig. 2B). A truncated form containing only the first 37 amino acids prior to the first Bateman domain interacted strongly with ␣2 and ␤2 (Fig. 2B), perhaps because the truncated form was better expressed (see Fig. 2B, right panel). However, the deletion of these 37 amino acids from the N terminus (a truncation at the start of the first Bateman domain) completely abolished the interaction with both ␣2 and ␤2 (Fig. 2B). Control experiments using the different LexA-␥2 constructs and the empty vector pACT2 gave negligible (Ͻ1 unit) ␤-galactosidase activity (not shown). Therefore, our results suggest that 37-47 amino acids immediately prior to the first Bateman domain of the three ␥ subunits are necessary for their interaction with ␣2 and ␤2.
Because yeast contains orthologues to the three AMPK subunits (AMPK␣, Snf1; AMPK␤, Gal83/Sip1/Sip2; and AMPK␥, Snf4), we studied whether the interaction between AMPK␤ and the three ␥ subunits was dependent on the presence of the orthologous Snf1/AMPK␣ subunit. With this aim, we repeated the two-hybrid experiments in yeast cells lacking the SNF1 gene (snf1⌬ mutant). As shown in Fig. 2C, the three ␥ subunits interacted with the ␤2 subunit in the absence of Snf1/AMPK␣. These results indicated that the ␤ and the ␥ subunits interacted directly. The lower levels of interaction observed in snf1⌬ mutants may suggest that the presence of the ␣ subunit stabilizes the ␤-␥ interaction.
Mutations in the ␥2 Bateman Domains Do Not Affect Binding to ␣2 and ␤2-Mutations in the PRKAG2 gene, encoding the ␥2 subunit, cause heart diseases of varying degrees of severity that appear to be caused by excessive glycogen storage (28,29). Several mutations have been described, e.g. R302Q (30), L-insert (an insertion of an extra Leu residue between the conserved Arg 350 -Glu 351 ) and H383R (31), T400N and N488I (28), R531G (32), and R531Q (33). In all cases, the described mutations affect critical residues in different CBS motifs of the ␥2 subunit and, with the possible exception of the L-insert mutation, produce proteins with deficient AMP binding capacity (8,33). We studied four of these mutations (R302Q, L-insert, H383R, and T400N (Fig.  3A)) and checked by two-hybrid analysis whether the mutated forms interacted properly with ␣2 and ␤2. Compared with the wild type, all of the mutants interacted normally with ␣2, and the interaction was increased in the absence of glucose, as reported previously for the wild type (10) (Fig. 3B). None of the mutations affected the two-hybrid interaction of ␥2 with ␤2 either (Fig. 3C).
Alignment of ␥ Subunit Sequences-We next aligned the sequences of human ␥1, ␥2, and ␥3 with those of the orthologues from Drosophila melanogaster, Dictyostelium discoideum, S. cerevisiae (Snf4), and Schizosaccharomyces pombe (Fig. 4). According to the UNIPROT data base, the first Bateman domains of ␥1, ␥2, and ␥3 begin at Lys 42 , Lys 279 , and FIGURE 2. Requirements for the interaction between ␥ and ␤ subunits using two-hybrid analysis. Yeast CTY10.5d strain was transformed with plasmids expressing GAD-AMPK␣2 or GAD-AMPK␤2 and either LexA-AMPK␥1 (wild type and N-terminal deleted form, Nt-⌬) (A) or LexA-AMPK␥2 (full-length, short form and N-terminal deleted form, Nt-⌬) (B). Transformants growing exponentially in SC-4% glucose medium were harvested, and the ␤-galactosidase activity was measured. Values correspond to means from four to six different transformants (S.D. Ͻ 15% in all cases; not shown). Crude extracts from these transformants were analyzed by Western blot using anti-LexA polyclonal antibodies. One representative transformant from each interaction is shown. C, interaction between ␥ and ␤ in snf1⌬ mutants. FY250 (wild type) and FY250 snf1⌬ mutant strains containing the pSH18 -18 reporter plasmid were transformed with plasmids expressing GAD-AMPK␤2 and either LexA-AMPK␥1, LexA-AMPK␥2, or LexA-AMPK␥3-D. Transformants growing exponentially in SC-4% glucose medium were analyzed as above. Bars indicated Ϯ S.D. Crude extracts from these transformants were analyzed by Western blot using anti-LexA polyclonal antibodies. One representative transformant from each interaction is shown.
Lys 202 , respectively. In a recent study, we expressed in bacteria constructs of ␥1, ␥2, and ␥3 that commenced near this point and showed that they bound two molecules of AMP or ATP with the expected affinities (8). Thus, upstream residues are not required for binding of the regulatory nucleotides. However, the alignment in Fig. 4 shows that there is a stretch of ϳ25 amino acids (which we shall call the "pre-CBS1 sequence") prior to the first Bateman domain that is well conserved across the three human ␥ isoforms and the four distantly related eukaryotic species. There is a conserved pattern of hydrophobic residues, whereas Phe, Tyr, and Asp residues at positions 12, 19, and 20 and a basic residue at position 16 (His, Lys, or Arg) are completely conserved.
Requirement for the Pre-CBS1 Sequence Revealed by Co-precipitation-To test the importance of this pre-CBS1 sequence for complex formation and also to map the interactions using a method different from two-hybrid analysis, we made constructs of human ␥1, ␥2, and ␥3 that were either truncated at the start of the first Bateman domain (␥1 S , ␥2 S , and ␥3 S ; S represents short) or that contained the 25-to 26-residue pre-CBS1 sequence shown in Fig. 4 (␥1 L , ␥2 L , and ␥3 L ; L represents long). These were co-expressed in HeLa cells with ␤1 and Myctagged ␣1, and the ␥ subunits were immunoprecipitated with anti-FLAG antibody, making use of the FLAG tag at the C termini of all ␥ variants. Fig. 5A shows the results of Western blotting using anti-Myc and anti-FLAG antibodies to detect the ␣ and ␥ subunits, respectively. A consistent finding was that Myc-␣1 (detected using anti-Myc antibody) only co-precipitated with ␥1, ␥2, and ␥3 when the latter contained the pre-CBS1 sequence (compare lane 1 (L variants) with lane 5 (S variants) for each ␥ subunit in the anti-Myc blot). Unfortunately, the ␥1 S variant appeared to be expressed at a much lower level than the ␥1 L variant. Also, for reasons that are unclear, it always migrated as a doublet, with one band migrating faster than ␥1 L as expected, but the other migrating more slowly. Nevertheless, the Myc-tagged ␣1 subunit (which was expressed equally well in both cases) only co-precipitated significantly with the ␥1 L variant (Fig. 5A, upper panels). Expression of the ␥2-FLAG fusions was low using either the long or short variant. Despite this, there was evidence that Myc-␣1 only co-precipitated with ␥2 L and not with ␥2 S (Fig. 5A, middle panels). The results were most clear-cut with the ␥3 variants (Fig. 5A, lower  panels). The ␥3-FLAG and Myc-␣1 fusions expressed equally well irrespective of the presence or absence of the pre-CBS1 sequence on ␥3, but Myc-␣1 only co-precipitated significantly with ␥3 L and not with ␥3 S . Very similar results were obtained when the constructs were expressed in HEK293 rather than HeLa cells (not shown). We also expressed Myc-␣1, ␤1, and the ␥-FLAG variants in both HeLa and HEK293 cells, immunoprecipitated the corresponding AMPK complexes using anti-Myc or anti-FLAG antibodies, and measured AMPK activity. Fig. 5B shows that active complexes were only obtained in anti-Myc immunoprecipitates from HEK293 cells using the long variants (␥1 L , ␥2 L , and ␥3 L ). The rather low level of activity obtained with ␥2 L was consistent with the generally low level of expression obtained using this construct compared with ␥1 L or ␥3 L (Fig. 5A). When the short variants (␥1 S , ␥2 S , and ␥3 S ) were expressed instead, the activities obtained were not significantly different from the background level in untransfected cells. Similar results were obtained by immunoprecipitation using anti-FLAG antibodies (not shown).
To confirm that AMPK complexes containing the pre-CBS1 sequence were regulated normally in intact cells, we also treated HEK293 cells expressing Myc-␣1, ␤1, and ␥1 L -or ␥1 S -FLAG with 10 mM deoxyglucose. Fig. 5C shows that the ␥1 L -FLAG complex was activated by deoxyglucose as expected, whereas only background levels of activity were observed in the Yeast CTY10.5d strain was transformed with plasmids expressing the indicated mutated forms of AMPK␥2 (LexA-AMPK␥2) and GAD-AMPK␣2. Transformants growing exponentially in SC-4% glucose medium were washed with water and shifted to SC-0.05% glucose medium for 3 h. Then samples were harvested from the cultures grown at high and low glucose conditions, and the ␤-galactosidase activity was measured. Values correspond to means from four to six different transformants (bars indicated Ϯ S.D.). C, interaction with AMPK␤2. Transformants growing exponentially in SC-4% glucose medium were harvested, and the ␤-galactosidase activity was measured. Values correspond to means from four to six different transformants (bars indicated Ϯ S.D.).
untransfected cells or cells expressing ␥1 S -FLAG, with or without deoxyglucose.
The Pre-CBS1 Sequence of ␥ Interacts with the ␤ Subunit-To test whether the pre-CBS1 sequence interacts with the ␣ or the ␤ subunits, we co-expressed the ␥1 S , ␥1 L , ␥2 S , ␥2 L , ␥3 S , or ␥3 L constructs in HeLa cells together with plasmids encoding either Myc-␣1, Myc-␣2, ␤1, or ␤2 alone, i.e. without DNA encoding the third subunit. Cell lysates were immunoprecipitated with anti-FLAG, and the respective ␣ or ␤ subunit was detected in the immunoprecipitates with the corresponding antibodies. The experiments with ␤1 or ␤2 (detected using a pan-␤ antibody) showed that both isoforms were able to form a complex with ␥1 L or ␥3 L , but not with ␥1 S or ␥3 S (Fig. 6A). We also attempted the experiment with the ␥2 constructs but, probably because of the low level of expression, any complex formed with ␥2 L was not detectable (not shown). On the other hand, Myc-␣1 or Myc-␣2 appeared to form complexes with ␥1 and ␥3 variants, irrespective of the presence or absence of the pre-CBS1 sequence (Fig. 6B). These results suggest that, although the interaction between the ␤ and ␥ subunits requires the pre-CBS1 sequence, ␣1 or ␣2 interact with ␥ subunits independently of this sequence, i.e. via the Bateman domains. Surprisingly, this is different from the results obtained in experiments when a ␤ subunit was also expressed (e.g. Figs. 5A and 6C), where the association of ␣ and ␥ is dependent on the presence of the pre-CBS1 sequence.
Recently, it has been proposed that the ␤ and ␥ subunits of AMPK do not interact directly, but only indirectly via the ␣ subunit (17). The results in Fig. 6A argue against that model, but a caveat was that HeLa cells do express low levels of endogenous ␣ subunits. To rule out the possibility that endogenous ␣ subunits could be providing a bridge between the overexpressed ␤ and ␥ subunits, we co-expressed the ␥-FLAG variants and ␤1, with or without Myc-␣1, in wild-type or double knock-out (␣1 Ϫ/Ϫ and ␣2 Ϫ/Ϫ ) mouse embryo fibroblasts (WT or KO MEFs). Unfortunately, the MEF cell lysates contain an abundant protein that migrates just behind the ␣ subunits on SDS-PAGE. This protein binds the anti-␣1/␣2 antibody and/or the second antibody used, and therefore almost obscures the ␣ subunits when cell lysates were analyzed by Western blotting. However, this protein was not present in the immunoprecipitates. Endogenous ␣ subunits from WT MEF cells could be seen to co-precipitate with ␥1 L (lane 8 in Fig. 6C), but no signal was obtained in the equivalent samples from KO MEFs (lane 8 in Fig. 6D), confirming the complete absence of ␣ subunits from KO cells. Recombinant Myc-␣1 was also observed to co-precipitate with ␥1 L when Myc-␣1 was expressed in either WT or KO MEFs (lane 4 in Fig. 6, C and  D). Despite the complication of the abundant protein in the MEF cell lysates, the results clearly show that ␤1 interacts with ␥1 L but not with ␥1 S , even in KO MEFs that completely lack both ␣ subunits (antipan-␤ blot, lane 8, Fig. 6, C and D). The presence of endogenous ␣1 in WT MEFs, or overexpressed Myc-␣1 in either WT or KO MEFs does, however, appear to markedly increase the amount of ␤1 that co-immunoprecipitates with ␥1 L but not with ␥1 S . These observations suggest that the ␣ subunit does stabilize ␤-␥ interaction, although it is not essential for the interaction to occur.
Both Bateman Domains of the ␥ Subunits Are Also Necessary to Allow the Formation of a Heterotrimeric Complex-We also made sequential deletions from the C terminus of ␥1 and found that, upon elimination of the last CBS motif of the second Bateman domain, the two-hybrid interactions with ␣2 and ␤2 were completely lost (Fig. 7). Successive deletions of the remaining CBS motifs gave the same negative results. Control experiments using LexA-␣2 or LexA-␤2 and the empty vector pACT2 gave negligible (Ͻ1 unit) ␤-galactosidase activity (not shown). Western blot analysis indicated that the truncated GAD-␥1 proteins were produced at similar levels (Fig. 7, right panel).
We also made various truncations of human ␥2 that contained only CBS motifs 3-4 (Bateman domain 2) or 2-4 (half Bateman domain 1 plus Bateman domain 2), as well as one that commenced at residue 244 (thus containing the pre-CBS1 sequence) but was truncated at the end of the first Bateman domain and did not contain the second domain. None of these constructs formed functional AMPK complexes when co-expressed with ␣1 and ␤1 (data not shown). These results indicated that the two Bateman domains may fold into a combined structure necessary to allow the formation of an active heterotrimeric complex. If this structure is altered, e.g. by deleting one of the CBS motifs, then the formation of a functional heterotrimeric complex is prevented.

DISCUSSION
AMPK exists as heterotrimeric complexes, and all three subunits are required to form a functional, AMP-activated complex when the full-length subunits are expressed (14,15). Because the crystal structure of the complex has not been determined yet, alternative methods have been used to obtain information about its architecture. For example, co-expression and co-immunoprecipitation techniques have been used to show that only the C-terminal domain of ␤ is required to form a functional complex with ␣ and ␥ (11), whereas additional C-ter- minal truncations have been used to more precisely map the regions of the C-terminal domain of ␤1 required for binding to ␣ (residues 186 -270) and ␥ (residues 246 -270) (16). The same technique was used to show that a C-terminal region of ␣1 lacking the last 75 residues (313-473) was all that was required for binding to ␤1 (16). Two-hybrid analysis has also been used as an alternative technique to analyze the interactions among the three different subunits (34,35). Recently, we reported that only the C-terminal domain of ␣2 (residues 313-552) was required for the interaction with ␥1 (10). The two-hybrid technique has also been used extensively to study the architecture of the yeast orthologue of mammalian AMPK, i.e. the SNF1 complex (36). The results are broadly similar to those obtained in the mammalian system: (i) yeast ␣ subunit (Snf1) interacts via its C-terminal domain with both the ␥ (Snf4) and ␤ (Gal83/ Sip1/Sip2) subunits (37); (ii) several deletions or point mutations in Snf1 also defined critical residues in the kinase FIGURE 5. A, co-precipitation of Myc-␣1 with ␥-FLAG constructs when coexpressed with ␤1 in HeLa cells; B, AMPK activity in anti-Myc immunoprecipitates when Myc-␣1 and ␤1 were co-expressed with ␥-FLAG constructs in HEK293 cells; C, same as for B but showing the effect of treatment of cells with deoxyglucose on the activity of the ␥1-FLAG constructs. In A, plasmids encoding Myc-␣1 and ␤1 were co-expressed in HeLa cells with ␥-FLAG fusions from ␥1, ␥2, or ␥3 that either contained (␥1 L , ␥2 L , and ␥3 L ) or did not contain (␥1 S , ␥2 S , and ␥3 S ) the pre-CBS1 sequence. Cell lysates were immunoprecipitated with anti-FLAG and lysates, and immunoprecipitates from transfected cells and untransfected controls (Ϫ) were analyzed by Western blotting using anti-Myc to detect recombinant ␣ and anti-FLAG to detect recombinant ␥. In B, plasmids encoding Myc-␣1 and ␤1 were co-expressed in HEK293 cells with the same set of ␥-FLAG variants. Extracts from transfected cells and untransfected controls (Ϫ) were subjected to immunoprecipitation with anti-Myc antibody and assayed for AMPK activity. In C, the experiment used the ␥1-FLAG constructs as in B, except that the cells were incubated with or without 10 mM deoxyglucose for 30 min prior to lysis, and the complexes were immunoprecipitated with anti-FLAG antibody. FIGURE 6. A, co-precipitation of ␤1 and ␤2 with ␥-FLAG variants in HeLa cells; B, co-precipitation of ␣1 and ␣2 with ␥-FLAG variants in HeLa cells; C, coprecipitation of ␤1 with ␥-FLAG variants in wild type MEFs; D, co-precipitation of ␤1 with ␥-FLAG variants in ␣1/␣2 double knock-out MEFs. In A, plasmids encoding ␤1 or ␤2 were co-expressed in HeLa cells with ␥-FLAG fusions with ␥1 or ␥3 that either contained (␥1 L and ␥3 L ) or did not contain (␥1 S and ␥3 S ) the pre-CBS1 sequence. Cell lysates were immunoprecipitated with anti-FLAG antibody and lysates (L) and immunoprecipitates (IP) from transfected cells were analyzed by Western blotting using anti-pan ␤ antibodies. In B, plasmids encoding Myc-␣1 and Myc-␣2 were co-expressed in HeLa cells with the same ␥-FLAG variants as in A. Cell lysates were immunoprecipitated with anti-FLAG antibody and lysates and immunoprecipitates from transfected cells and untransfected controls were analyzed by Western blotting using anti-Myc (to detect ␣ subunits) and anti-FLAG (to detect ␥ subunits). In C, plasmids encoding Myc-␣1 and/or ␤1 were co-expressed in WT MEFs with ␥1-FLAG fusions that either contained (␥1 L ) or did not contain (␥1 S ) the pre-CBS1 sequence. Cell lysates were immunoprecipitated with anti-FLAG antibody and lysates and immunoprecipitates from transfected cells, and untransfected controls were analyzed by Western blotting using a mixture of anti-␣1 and anti-␣2 antibodies or anti-pan-␤ antibodies. D, the same as for C except that double knock-out MEFs were used.
domain and C-terminal domain involved in the interaction with Snf4 (38); (iii) yeast ␤ subunits contain C-terminal domains referred to as the ASC (association with Snf1 complex) domain reported to be involved in the interaction with Snf4 (␥ subunit) and a central KIS (kinase interacting sequence) domain, which overlaps with the glycogen-binding domain of mammalian ␤ subunits and is reported to interact with the ␣ subunit (Snf1) (37,39).
Until now, nothing has been known in either the yeast or the mammalian system about which part of the ␥ subunit is involved in the interaction with the ␣ and ␤. In this study we present evidence that a conserved region of 20 -25 amino acids immediately N-terminal to the first Bateman domain of mammalian ␥ subunits is required for binding to the ␤ subunits. Our results are based on analysis of the interactions using two-hybrid analyses of mammalian subunits expressed in yeast, as well as by co-precipitation and measurement of kinase activity with ␣, ␤, and ␥ subunits co-expressed in mammalian cells. The results obtained with the expression of truncated forms of ␥1, ␥2, and ␥3 in mammalian cells show that a region of 20 -25 amino acids immediately prior to the first Bateman domain is essential for formation of a functional ␣␤␥ complex (Fig. 5). This sequence is particularly well conserved in a central region of 17 amino acids, which is rather hydrophobic and has the consensus sequence SXIYMKFMRSHKCYDLI. The sequence contains three residues that are completely conserved between human ␥1, ␥2, and ␥3 and their orthologues in D. melanogaster, D. discoideum, S. cerevisiae, and S. pombe, i.e. a Phe at position 7 and a Tyr-Asp doublet at position 14 -15 (in bold in the consensus sequence). There is also a conserved basic residue at position 11, and hydrophobic residues are conserved at positions 4, 8, 16, and 17 (underlined in the consensus sequence) (Fig. 4). Although our results suggest that this sequence is required for the interaction between ␥ and ␤ subunits, at present, we cannot rule out the possibility that the removal of this sequence causes major conformational changes in the overall structure of the ␥ subunits that precludes their interaction with the ␤ subunits.
Additionally, we have also used the two-hybrid method to analyze the interaction with ␣ and ␤ of several point mutations in ␥2 that caused a hereditary heart disease in humans (Wolf-Parkinson-White syndrome with cardiac hypertrophy). Our results indicate that the pathogenic defect associated with these mutations is not due to deficiencies in binding of ␥2 to ␣2 and ␤2. This is consistent with findings that these ␥2 mutations give rise to complexes that are active, although in some cases defective in AMP activation, when co-expressed with ␣ and ␤ subunits in mammalian cells (8,33,40). The findings that these mutant complexes were active suggested that they contained all three subunits, although this had not been directly addressed in the previous studies. Recently, evidence was presented, using a co-precipitation approach, that mouse ␤2 and ␥1 subunits do not interact directly (17). The authors proposed instead that the ␣ subunit bridges the interaction between ␤ and ␥. Our results do not support this model. First, yeast two-hybrid analysis indicates that the interaction between ␤ and ␥ subunits occurs in the absence of Snf1/AMPK␣ subunit (Fig. 2C). Second, both ␤1 and ␤2 interacted with ␥1 and ␥3 when overexpressed together in HeLa cells without an ␣ subunit, and this was dependent on the pre-CBS1 sequence of the ␥ subunits, because it did not occur with the short variants (Fig. 6A). Both ␣1 and ␣2 also interacted with ␥1 and ␥3 in the absence of a co-expressed ␤ subunit, but this was not dependent on the pre-CBS1 sequence, because the interaction occurred equally well with the short as the long ␥ variants (Fig. 6B). The fact that the ␤ subunits require the pre-CBS1 sequence to interact with the ␥ subunits, whereas the ␣ subunits do not, argues against the idea that ␣ subunits provide a bridge between ␤ and ␥. One caveat with our experiments in HeLa cells is that these cells do express low level of endogenous ␣ subunit. To rule out the possibility that the endogenous ␣ subunit might bridge the ␤-␥ interaction, we repeated the experiment in WT MEFs and double knock-out cells that completely lacked ␣ subunits (KO MEFs). The results clearly showed that a ␤-␥ interaction, which was dependent on the pre-CBS1 sequence, was still evident in the KO MEFs. However, the presence of ␣ subunits, either endogenous in the WT MEFs or recombinant Myc-␣1 subunit in WT or KO MEFs, did appear to greatly increase the amount of ␤1 recovered in the ␥1-FLAG immunoprecipitates (Fig. 6, C and D). Thus, the presence of an ␣ subunit may stabilize the ␤-␥ interaction. The lower levels of interaction observed between ␤ and ␥ in snf1⌬ mutants are also consistent with this hypothesis.
Some rather surprising findings were that when ␣ and ␥ subunits were expressed on their own, they interacted even in the absence of the pre-CBS1 sequence (Fig. 6B), but when they were co-expressed with a ␤ subunit, co-precipitation with the ␥ subunits was now dependent on the presence of the pre-CBS1 sequence on ␥ (Figs. 5A and 6C). Our interpretation of these somewhat puzzling findings is that the ␣ and ␤ subunits may form a dimeric complex that interacts differently with the ␥ subunits than when the ␣ subunit is present on its own. . The second Bateman domain in ␥1 is required to interact with ␣2 and ␤2. Yeast CTY10.5d strain was transformed with plasmids expressing LexA-AMPK␣2 or LexA-AMPK␤2 and C-terminal truncated forms of GAD-AMPK␥1. Transformants growing exponentially in SC-4% glucose medium were harvested, and the ␤-galactosidase activity measured. Values correspond to means from four to six different transformants (S.D. Ͻ 15% in all cases; not shown). Crude extracts from these transformants were analyzed by Western blot using anti-HA polyclonal antibodies (plasmid pACT2-generated GAD-HA fusion proteins). One representative transformant from each interaction is shown.
We have also made sequential deletions from the C terminus of ␥1 and found that, upon elimination of even one of the four CBS motifs, the two-hybrid interactions with ␣2 and ␤2 were completely lost. In agreement with this observation, we also found that a C-terminal deletion of ␥2 (lacking the second Bateman domain) was unable to form an active heterotrimeric complex. One plausible interpretation of these results is that both Bateman domains are required to form a functionally combined structure. If one of the CBS motifs or Bateman domains is deleted, then the structure does not fold properly, leading to an altered conformation of the protein that cannot bind to ␣ and ␤ even though the pre-CBS1 sequence is present.
Addendum-At the time that the manuscript was under revision, Townley and Shapiro (41) have defined the crystal structure of the AMPK complex from the yeast S. pombe. They demonstrate that the ␥ and ␤ subunits interact directly and that the pre-CBS1 sequence of the ␥ subunit participates in the binding to the ␤ subunit.